Searcharxiv⌕ Search

arXiv subjects

Alexander Conway

Publications and source records attributed to Alexander Conway.

7 recordsLinked to original sources
arXiv2025

syftr: Pareto-Optimal Generative AI

Retrieval-Augmented Generation (RAG) pipelines are central to applying large language models (LLMs) to proprietary or dynamic data. However, building effective RAG flows is complex, requiring careful selection among vector databases, embedding models, text splitters, retrievers, and synthesizing LLMs. The challenge deepens with the rise of agentic paradigms. Modules like verifiers, rewriters, and rerankers-each with intricate hyperparameter dependencies have to be carefully tuned. Balancing tradeoffs between latency, accuracy, and cost becomes increasingly difficult in performance-sensitive applications. We introduce syftr, a framework that performs efficient multi-objective search over a broad space of agentic and non-agentic RAG configurations. Using Bayesian Optimization, syftr discovers Pareto-optimal flows that jointly optimize task accuracy and cost. A novel early-stopping mechanism further improves efficiency by pruning clearly suboptimal candidates. Across multiple RAG benchmarks, syftr finds flows which are on average approximately 9 times cheaper while preserving most of the accuracy of the most accurate flows on the Pareto-frontier. Furthermore, syftr's ability to design and optimize allows integrating new modules, making it even easier and faster to realize high-performing generative AI pipelines.

cs.AI↗
arXiv2015

Weyl group orbits on Kac--Moody root systems

Let $\mathcal{D}$ be a Dynkin diagram and let $Π=\{α_1,\dots ,α_{\ell}\}$ be the simple roots of the corresponding Kac--Moody root system. Let $\mathfrak{h}$ denote the Cartan subalgebra, let $W$ denote the Weyl group and let $Δ$ denote the set of all roots. The action of $W$ on $\mathfrak{h}$, and hence on $Δ$, is the discretization of the action of the Kac--Moody algebra. Understanding the orbit structure of $W$ on $Δ$ is crucial for many physical applications. We show that for $i\neq j$, the simple roots $α_i$ and $α_j$ are in the same $W$--orbit if and only if vertices $i$ and $j$ in the Dynkin diagram corresponding to $α_i$ and $α_j$ are connected by a path consisting only of single edges. We introduce the notion of `the Cayley graph $\mathcal{P}$ of the Weyl group action on real roots' whose connected components are in one-to-one correspondence with the disjoint orbits of $W$. For a symmetric hyperbolic generalized Cartan matrix $A$ of rank $\geq 4$ we prove that any 2 real roots of the same length lie in the same $W$--orbit. We show that if the generalized Cartan matrix $A$ contains zeros, then there are simple roots that are stabilized by simple root reflections in $W$, that is, $W$ does not act simply transitively on real roots. We give sufficient conditions in terms of the generalized Cartan matrix $A$ (equivalently ${\mathcal D}$) for $W$ to stabilize a real root. Using symmetry properties of the imaginary light cone in the hyperbolic case, we deduce that the number of $W$--orbits on imaginary roots on a hyperboloid of fixed radius is bounded above by the number of root lattice points on the hyperboloid that intersect the closure of the fundamental region for $W$.

math.GR↗
arXiv2014

Measuring the Higgs Self-Coupling Constant at a Multi-TeV Muon Collider

A lepton collider in the multi-TeV range has the potential to measure the trilinear Higgs self-coupling constant $λ_{hhh}$ via the W-fusion mode $\ell^+\ell^- \rightarrow ν_\ell \barν_\ell h h$. In this paper we do a generator-level study to explore how center-of-mass energy spread, cone size, tracking resolution, and collision energy range affect how precisely a muon collider can measure $λ_{hhh}$ in comparison to an $e^+e^-$ collider. The smaller spread in center-of-mass energy and higher energy range of a muon collider improve cross section while the larger cone required to reduce beam-induced background hinders detection of double-Higgs events. Our results motivate a more detailed study of a multi-TeV muon collider and innovative detector and analysis technologies required for background rejection and precision measurement.

hep-ex↗
arXiv2014

Activity Modulation of Motor and Somatosensory Neurons in Learning

The cortical processes involved in learning are not well understood. Recent experiments have studied population-level response in the orofacial somatosensory (S1) and motor (S1) cortices of rhesus macaque monkeys during adaptation to a simple tongue protrusion task within and across multiple learning sessions. Initial findings have suggested the formation of cell assemblies during adaptation. In this report we explore differences in cell activity between successful and failed trials as the monkey learns during two sessions. The ability to directly compare data across multiple sessions is fairly new and until now research has mostly focused on the activity of neurons during successful trials only. We confirm findings of the development of coherently active cell assemblies and find that neural response differentiates significantly between successful and unsuccessful trials, particularly as the monkey adapts to the task. Our findings motivate further research into the differences in activity between successful and unsuccessful trials in these experiments.

q-bio.NC↗
arXiv2013

The Case for a Muon Collider Higgs Factory

We propose the construction of a compact Muon Collider Higgs Factory. Such a machine can produce up to \sim 14,000 at 8\times 10^{31} cm^-2 sec^-1 clean Higgs events per year, enabling the most precise possible measurement of the mass, width and Higgs-Yukawa coupling constants.

hep-ph↗
arXiv2013

Higgs Measurements at a Muon Collider

In light of the recent discovery of an approximately 126 GeV Higgs boson at the LHC, the particle physics community is beginning to explore the possibilities for a next-generation Higgs factory particle accelerator. In this report we study the s-channel resonant Higgs boson production and Standard Model backgrounds at a proposed μ+μ- collider Higgs factory operating at center-of-mass energy sqrt(s) = M_H with a beam width of 4.2 MeV. We study PYTHIA-generated Standard Model Higgs and background events at the generator level to identify and evaluate important channels for discovery and measurement of the Higgs mass, width, and branching ratios. We find that the H^0 -> bb and H^0 -> WW^* channels are the most useful for locating the Higgs peak. With an integrated luminosity of 1 fb^-1 we can measure a 126 GeV Standard Model Higgs mass accurately to within 0.25 MeV and its total width to within 0.45 MeV. Our results demonstrate the value of the high Higgs cross section and narrow beam resolution potentially achievable at a muon collider.

hep-ex↗